c3ttp

Small HTTP/1.1 client and server library for C3, built on std::net::tcp.

• c3ttp.c3i: public API

• src/: implementation

• test/: package tests

• examples/: usage examples

• HTTP/1.1 request and response parsing

• Content-Length and Transfer-Encoding: chunked

• buffered and streaming body APIs

• basic TCP client and server helpers

• keep-alive support for HTTP/1.1

• Linux epoll event loop for server workloads

• Linux prefork serving with SO_REUSEPORT

• TCP_NODELAY enabled for client connections and accepted server sockets

c3c compile-test .
(cd examples/hello_server && c3c build)
(cd examples/client_get && c3c build)
(cd examples/router_simple && c3c build)

For long-lived servers, do not initialize Server with tmem. Use mem or another non-temp allocator so per-request allocations can be reclaimed normally across many requests.

See c3ttp.c3i for the supported public surface.

• Request / Response
• Client.send / Client.send_url
• Server.listen / Server.serve_once / Server.serve
• Server.serve_evented / Server.serve_prefork
• read_request / read_response
• write_request / write_response

HTTP messages are parsed and serialized using strict HTTP/1.1 syntax and framing. On Linux, serve_evented uses nonblocking persistent connections. Its per-connection input and output buffers are allocated lazily and start at 1 KiB and 256 bytes respectively, growing automatically when needed.

Connection and prefork memory usage can be configured explicitly:

ServerOptions options = c3ttp::default_server_options();
options.max_connections = 512; // Default: 1024; zero means unlimited.
Server server;
server.init(mem, "127.0.0.1", 8080, 64, options);
server.serve_prefork(&handler, 4)!!; // Explicit worker count; zero uses CPU count.

max_connections applies to each event loop, so in prefork mode the process-wide theoretical maximum is the per-worker limit multiplied by the worker count.